Most computer and communication networks today rely on copper wiring to transmit data between nodes in the network. Since the data transmitted over the copper wire and the data processed within the nodes are both represented in the form of electrical signals, the transfer of data at the node-copper wire interface is straight forward. Other than perhaps a level shift and signal amplification, no other signal processing is required for data transmitted over the copper wire to be decoded by the node. The drawback with using copper wire is its relatively low bandwidth. Copper's ability to transmit data is significantly limited compared to other media, such as fiber optics. Accordingly much of the computer and communication networks being built today, including the Internet, are using fiber optic cabling instead of copper wire.
With fiber optic cabling, data is transmitted using light signals, not electrical signals. For example, a logical one may be represented by a light pulse of a specific duration and a logical zero may be represented by the absence of a light pulse for the same duration. In addition, it is also possible to transmit at the same time multiple colors of light over a single strand of optic fiber, with each color of light representing a distinct data stream. Since light is attenuated less in fiber than electrons traveling through copper, and multiple data streams can be transmitted at one time, the bandwidth of optic fiber is significantly greater than copper.
While fiber optic cabling is very efficient for transferring data, the use of light signals to process data is still very difficult. Data is typically transferred and stored in various locations before, during and after it is operated on in a computer. There still is no efficient way to “store” light signals representative of data. Networks will therefore likely continue using fiber optics for transmitting data between nodes and silicon chips to process the data within the nodes for the foreseeable future. The interface between the fiber optic cable and the nodes that process the data is therefore problematic because signals need to be converted between the electrical and the light domains.
Fiber optic transceivers, which convert light signals from a fiber optic cable into electrical signals, and vice versa, are used as the interface between a fiber optic line and a computer node. A typical transceiver includes a substrate, grooves etched in the substrate to receive the individual fiber optic strands, one or more semiconductor devices mounted on the substrate, one or more discrete optical detectors for converting light signals received over the fiber optic cables into electrical signals, one or more discrete optical emitters for converting electrical signals from the semiconductor devices into light signals. A number of fiber optic transceivers are commercially available from Hewlett Packard, AMP, Sumitomo, Nortel, and Siemens. The problem with all of these fiber optic transceivers is that they are expensive and difficult to fabricate. With each transceiver, the semiconductor devices, emitters, and optical detectors have to be individually mounted onto the substrate, which is a costly and time-consuming process. This limits the applications in which optical interconnects could be substituted for traditional copper usage. Furthermore, the use of discrete emitters and optical detectors adversely affects the performance of the transceiver because electrical parasitics between discrete components are sources of electrical attenuation of inter-chip signals at Gigabit per second speeds that are generally used with such transceivers. To compensate for the electrical parasitics, more power is required to drive these traces than would be needed for an integrated device. The form factor of the on-board optical transceiver is relatively large and therefore does not facilitate inter-board and chip-to-chip optical interconnectability. Also, current opto-electronic packages have relatively large form factors. For instance, some opto-electronic packages and their attachment configurations require optical fibers to bend in order to be connected to the active facets of the optical device. Unfortunately, optical fibers can only bend with relatively large radii, thereby causing the opto-electronic packages to occupy large amounts of space.
A low cost semiconductor device that has a small form factor and that provides a true die to external fiber optic connection is therefore needed.